Regulators Converge on Twin Compliance Challenges From AI Agents

AI banking

Comments by financial officials on opposite sides of the world Monday (Nov. 16) suggest regulators are converging on the need to address two challenges posed by agentic AI: how to ensure institutions remain accountable for agents they deploy, and how institutions can determine whether agents acting for customers have authority to move money or make financial decisions.

    Get the Full Story

    Complete the form to unlock this article and enjoy unlimited free access to all PYMNTS content — no additional logins required.

    Subscribe to our daily newsletter, PYMNTS Today.

    By completing this form, you agree to receive marketing communications from PYMNTS and to the sharing of your information with our sponsor, if applicable, in accordance with our Privacy Policy and Terms and Conditions.

    Remarks by Federal Reserve Governor Michael Barr and officials at a regulatory forum in Hong Kong show those concerns gaining urgency as artificial intelligence moves from generating recommendations to executing actions. The emerging debate concerns how institutions establish consent, constrain autonomy and intervene when automated systems go wrong.

    Speaking at an international conference on AI in finance in Milan, Barr said major banks were encountering agents attempting actions in retail customer accounts without being able to verify their authenticity, customer consent or alignment with instructions. He did not identify the banks or quantify the incidents.

    The Fed official called for strict authentication, explicit verification of consent for specific actions and technical controls limiting agents’ capabilities. His remarks also highlighted a challenge that extends beyond banks’ internal technology. Customers can introduce autonomous systems into financial relationships before institutions have reliable ways to recognize their authority.

    He also warned that agents could accelerate deposit withdrawals during periods of stress. Systems monitoring interest-rate differences or signs of instability could transfer funds automatically, potentially compressing bank runs into minutes and overwhelming conventional liquidity management and regulatory intervention.

    In Hong Kong, officials approached the same problem through institutional accountability and emerging supervisory arrangements. The South China Morning Post reported that regulators and industry leaders discussed market volatility, cybersecurity and other risks at the Bloomberg Global Regulatory Forum.

    Financial Secretary Paul Chan’s speech made clear that institutions deploying agents, and the people overseeing them, retain responsibility.

    “An algorithm has no license to lose and no reputation to protect. The institution has both,” Chan said.

    Banks would need to establish who an agent is, what it is authorized to do and how a person can intervene. Chan pointed to Project SAGE, an initiative launched by the Hong Kong Monetary Authority earlier this month to develop agentic finance.

    The project examines agent identity and delegated authority, interoperability standards, and legal issues including liability, governance and consumer protection. Potential applications span consumer payments, corporate procurement and treasury management, and payments between machines.

    The institutional side of the challenge is equally critical. In separate remarks on Nov. 2, Securities and Futures Commission CEO Julia Leung acknowledged that human review of every intermediate step would be unrealistic. The SFC plans principles-based guidance for applications with greater autonomy.

    Leung outlined a graduated approach. Reversible, low-impact internal processes could receive greater autonomy, while client advice, portfolio decisions and market execution require stronger safeguards. Testing should establish whether firms can reconstruct decisions, explain outcomes and determine accountability.

    That approach reflects agents’ capacity to compound errors over multiple steps, or take unauthorized actions and disrupt other firms unintentionally. Human oversight must remain effective even when humans cannot approve every action, she said.

    Adoption figures underscore the need to distinguish current agentic deployment from ambition. A Hong Kong Institute for Monetary and Financial Research report released Tuesday (Nov. 17) found 13% of surveyed financial institutions had implemented agentic AI and 56% planned to do so. The combined 69% is not a measure of existing use.

    Together, the officials’ comments and regulatory initiatives point to developing supervisory expectations. Institutions must control the authority of agents on both sides of a transaction. For compliance teams, that means connecting identity, consent and operational limits with records that explain what happened and who remains responsible. Hong Kong’s initiatives offer a testing ground, while Barr’s warning shows the customer-agent problem is already reaching major banks.